Hyperspectral images (HSIs) are often corrupted by a mixture of several typesof noise during the acquisition process, e.g., Gaussian noise, impulse noise,dead lines, stripes, and many others. Such complex noise could degrade thequality of the acquired HSIs, limiting the precision of the subsequentprocessing. In this paper, we present a novel tensor-based HSI restorationapproach by fully identifying the intrinsic structures of the clean HSI partand the mixed noise part respectively. Specifically, for the clean HSI part, weuse tensor Tucker decomposition to describe the global correlation among allbands, and an anisotropic spatial-spectral total variation (SSTV)regularization to characterize the piecewise smooth structure in both spatialand spectral domains. For the mixed noise part, we adopt the $\ell_1$ normregularization to detect the sparse noise, including stripes, impulse noise,and dead pixels. Despite that TV regulariztion has the ability of removingGaussian noise, the Frobenius norm term is further used to model heavy Gaussiannoise for some real-world scenarios. Then, we develop an efficient algorithmfor solving the resulting optimization problem by using the augmented Lagrangemultiplier (ALM) method. Finally, extensive experiments on simulated andreal-world noise HSIs are carried out to demonstrate the superiority of theproposed method over the existing state-of-the-art ones.
展开▼
